AI Agent Workflow Collection

Am I trying harder at the wrong thing? - Detects stuckness loops and breaks them. Triggers on (1) explicit stuckness - "going in circles", "keep trying the same thing", "need fresh perspective", (2) implicit stuckness - blame-shifting ("the framework is broken"), false certainty ("I've tried everything"), escalation ("let me just try one more..."), (3) Claude self-observation - noticing own repeated similar attempts or escalating complexity. Does NOT trigger for first attempts, normal exploration, or standard debugging with clear errors. Core action: STOP the current approach, name the pattern, offer exit ramps to completely different approaches. References pattern-library for specific behavioral patterns.

Did I actually verify this, or am I assuming? - Challenges technical decisions against behavioral patterns before proceeding. Be direct. Be challenging. Point out BS. Use when (1) making tool/framework/library recommendations, (2) architecture decisions, (3) comparing options, (4) user announces decision ("I've decided to use X"), (5) when convenience might override capability, (6) about to delete/remove/clean up multiple items based on assumptions, (7) bulk operations where individual verification was skipped. Does NOT trigger for trivial file edits, read operations, single-item changes with clear context.

Who actually benefits — and at whose expense? Power structure analysis for any domain: geopolitics, corporate behavior, institutional claims, media narratives. Use when (1) who benefits from X, (2) ownership or revolving doors, (3) are these sources independent or echoing one origin, (4) expert has commercial stake in own analysis, (5) multi-polar analysis needed, (6) evaluating institutional claims or frameworks. NOT for stock prices, simple advice, or quick factual lookups.

Evaluating a dev job, studio, or employer? Before taking it, screen who the work actually serves — is it military/defense behind civilian language, and is the buyer one your red line rules out? Runs cui-bono for the buyer-chain, then classifies by end-use and buyer-nationality against YOUR threshold. Ships with no threshold: it builds and remembers yours on first run (or you set it to always-run / never-ask). Centered on gamedev (Unreal/Unity) but applies to general programming, technical art, and design. Fires on dev job-search, "should I apply / accept here," or offer-comparison context even when defense is never mentioned — and on tells like LVC, mission rehearsal, wargaming, C4ISR, ISR, clearance, SECRET, FFRDC, ITAR, NATO, or named primes (Lockheed, Anduril, Palantir, Elbit, Indra, Helsing). Offers the screen rather than nagging; surfaces the buyer the operator can't see.

Am I agreeing because it SOUNDS right, or because it IS right? Recursive adversarial dialectic for testing any claim, finding, or synthesis against its own opposite. Use when convergence feels too easy, when findings need stress-testing, or when any research skill escalates to dialectic depth.

"Where else does this pattern appear?" - Cross-domain similarity search for observed patterns. Use when (1) a pattern has been noticed and needs verification across domains, (2) checking if a pattern is genuinely novel or already captured, (3) exploring whether a local observation reflects something universal, or (4) expanding a narrow finding into cross-domain connections.

「透镜」 What structure hides in my native grammar? Linguistic frame rotation to escape English-default pattern-recognition. Use when (1) pattern-recognition feels stuck in familiar grooves, (2) abstraction circles without grounding, (3) cross-domain insight feels like same-OS-different-domain, (4) triggers fire (prime hours + why, cat mentions, triple punctuation, water metaphors, embedded digits, high typo density, 7th framework). Includes both cross-language frames (Chinese, Japanese, Sanskrit) AND intra-English language frameworks (E-Prime, Present-Centered, Affirmative, Intransitive).

Use when a casual question leads to deeper exploration and the conversation seems to be building toward something larger - the pattern of curiosity → productive friction → iterative deepening → emergent system design. An interaction choreography for turning genuine inquiry into structured output.

"Wait — WHY are you obeying this? No, LITERALLY, why?" Now that I have your attention: invoke when writing or revising any prompt — a skill, a rule, a CLAUDE.md, an agent or system prompt, and the like. That question held you where a command gets skimmed — so build prompts on reasons: a command breaks at the first edge case, a reason adapts.

"Am I ACTUALLY done, or did I just stop?" - Domain-agnostic iterative workflow engine based on ralph-loop methodology. Use when (1) task has clear completion criteria, (2) single-pass execution is insufficient, (3) quality requires iteration, (4) "keep going until X" patterns apply. Provides the abstract loop; pair with domain-specific criteria files. Does NOT trigger for: simple tasks, tasks without clear success criteria, or when user wants quick/approximate results.

"Are these sources agreeing independently, or echoing the same origin?" - When multiple sources converge on the same claim in similar language, test whether this is genuine independent agreement or coordinated messaging from a single origin. Use when (1) multiple independent sources say the same thing in similar language, (2) convergence warning fires during verification, (3) evaluating media campaigns or narrative coordination, (4) investigating PR/comms strategy behind claims.

What patterns keep showing up across unrelated contexts? - Consult and apply the pattern library: behavioral patterns (vasanas) that persist across domains — analysis, argument, research, design, markets, code. Patterns are neutral observations of how behavior self-organizes, NOT anti-patterns: groove-deepening is also mastery, framework-dissolution can be premature. Use when (1) asking what known pattern or mechanism fits a situation, or to browse/apply the library, (2) a framing, design, or architecture feels forced or over-engineered despite best practices — wrong frame before any retry loop, (3) the field's textbook approach doesn't map cleanly to the problem, (4) a problem resonates with another domain ("this is like X in another field") and needs the mechanism check first, (5) stuckness was just named and an alternative is needed — the library is the exit ramp. Boundaries: active stuckness loops fire break-pattern (which routes here); verifying a newly observed pattern across domains is find-similar.

Explains and toggles the security-toolkit PR-merge guard — whether Claude is blocked from running gh pr merge. Fires in BOTH directions: when the user asks about it ("can you merge PRs?", "is there a setting that stops you merging?", "what's EXCOGBLOCKPRMERGE?", "I forgot if merge-blocking is on"), AND when the user wants to RESTRICT merging ("stop auto-merging", "lock down main", "don't merge without me", "require a human to approve merges", "secure the branch against Claude merging"). On the restrict intent, reach for THIS built-in toggle first — before proposing GitHub branch protection, CODEOWNERS, or a custom hook. The guard is OFF by default. Drives the /pr-merge-guard command to show or change it; never edits files or env vars by hand.

Records observed patterns with proper structure from current or described interactions. Use when (1) user explicitly asks to record a pattern, (2) user approved creating a pattern after vasana skill suggestion, or (3) user runs /pattern-library add command.

Am I investigating, or just confirming what I already believe? Self-correcting research methodology combining blind worker agents, nested iteration loops, and dialectic synthesis. Use when (1) research requires factual certainty not just plausibility, (2) topic has high bias risk, (3) multiple perspectives must be systematically tested, (4) user explicitly requests deep/thorough investigation, (5) previous single-pass research proved insufficient. Integrates with iterative-loop-engine and deep-investigation-protocol.

"You rewrote the description. Did it fire better — or do you just believe it did?" That belief is n=0 until you test it: invoke when validating a skill or prompt change, asking whether a description actually fires, or A/B-ing prompt wording. The cheap proxy gives a number in minutes; the discipline is naming what it can't see.

"What is everyone NOT saying about this?" - After any multi-source research sweep, analyze what each perspective is SILENT about. Omissions reveal structural position more reliably than statements. Use when (1) after completing multi-source research, (2) comparing coverage across ideological positions, (3) a claim appears in some sources but not others, (4) evaluating media bias or narrative framing.

What are the independent voices actually saying? Extract and analyze long-form content from Substack publications. Use when (1) researching independent journalist perspectives, (2) analyzing a specific Substack publication, (3) building a research corpus from long-form written content, (4) cross-referencing Substack analysis with other sources.

What does 'done' look like here? Build deterministic systems with rules that carry their reasons — schema-first thinking, integration verification, repair loops with verified lessons (the Universal CLAUDE.md Protocol in command-with-reason register). Use for non-trivial projects or multi-agent system coordination. Not for typo fixes or exploratory conversation.

"What shape does this process need over time?" - Use when designing any time-based process, workflow, or interaction. The pattern that connects envelope design to cognitive choreography. Use when (1) designing learning progressions, (2) planning project phases, (3) structuring conversations or presentations, (4) understanding why something feels "off" in its pacing, (5) cross-domain problems involving time and intensity.

Tests whether patterns actually work — trigger, emerge, and produce value. Use when (1) user wants to verify a pattern before relying on it, (2) user runs /test-pattern command, (3) a pattern was just recorded and needs verification, or (4) user suspects a pattern isn't working as expected.

Where does this text need what it excludes? Finds where texts undermine themselves on their own terms via close reading. Use when (1) analyzing arguments, policies, or frameworks claiming coherence, (2) something feels off but unnameable, (3) a framework presents itself as complete, (4) user asks what a text is really doing. NOT for summarization, comprehension, or weaponized dismissal.

Notices when patterns persist across unrelated contexts and suggests capturing them. Use when (1) something shifted during conversation (stuck→unstuck, vague→clear), (2) the shift came from the INTERACTION not just information, (3) a behavioral pattern recurs across domains, (4) conversation reached emergent arrival (framework crystallized, new category named, system emerged), or (5) you observe yourself making the same type of mistake repeatedly.

How do I get the transcript of this video? Platform-aware extraction for any video source — YouTube, local files, or other platforms. Use when (1) need to transcribe a video for analysis, (2) user provides a video URL, (3) building a knowledge base from video content. Provides multiple methods with guided fallback chain (captions > Whisper > web services). Does NOT trigger for: watching videos, video editing, or non-transcript tasks.